Feature Engineering Pipelines and Data Versioning Quiz

Answers and explanations appear as you take the quiz.

1. What is the primary purpose of a feature engineering pipeline?
   • To store trained model weights for later deployment
   • To schedule GPU jobs across a cluster
   • To transform raw data into the structured features a model consumes
   • To encrypt datasets at rest
2. Why compare a dataset’s content hash instead of relying only on its filename?
   • A hash guarantees every record is accurate
   • The filename can stay the same while its stored contents change
   • A filename contains all transformations
   • Hashing removes access restrictions
3. A feature pipeline is commonly represented as which kind of structure?
   • A single flat list of SQL queries that must all rerun every time
   • A relational database table
   • A neural network with hidden layers
   • A directed acyclic graph (DAG) of transformation steps
4. A model behaves differently after a data update. What can version history help investigate?
   • Which data and transformation versions were used, with other runtime factors checked separately
   • A guaranteed identical rerun without environment records
   • Proof the update was authorized
   • A universal legal-compliance certificate
5. Which of these is an example tool specifically used for feature stores or data versioning?
   • Photoshop
   • Microsoft Word
   • FFmpeg
   • Feast (or DVC / LakeFS)
6. Why is it important that the same feature logic runs during both training and production?
   • To reduce the cost of cloud storage
   • To make the model larger and more accurate automatically
   • To ensure the model sees consistently computed features and avoids training/serving mismatches
   • Because production servers cannot run Python
7. A team has the same data and code but a changed random state. What remains possible?
   • Different computed outputs despite matching versions
   • Versioning forces every random choice to match
   • The dataset filename must change
   • All models become deterministic
8. A reviewer asks which input record and transform produced a feature at a particular time. What record addresses that question?
   • Only the final prediction text
   • Feature lineage with source, transform version and timestamp
   • Only the dataset filename
   • Only a model parameter count